Motivation

In modern society, fitness has become an integral part of many people’s daily lives. However, there is limited in-depth data analysis on the behavioral patterns, demographic characteristics, and geographic distributions of gym-goers. This project aims to address this gap by systematically analyzing data on calorie expenditure, workout duration, and subscription plan choices among gym-goers to identify key factors influencing these behaviors. Our goal is to provide valuable insights for the fitness industry, public health researchers, and policymakers, enabling them to better understand fitness trends and develop more effective interventions and strategies to promote overall health.


Initial Questions

  • What factors contribute to the calories a user burns in gyms?

  • What factors contribute to the workout duration and frequency of users’ gym visits?

  • What factors contribute to the selection of users’ subscription plans?


Data Source and Cleaning

You can find the raw datasets here. Click the button to show relative code.

# Import 4 raw datasets
users_data = read_csv("datasets/users_data.csv")
gyms_data = read_csv("datasets/gym_locations_data.csv")
history_data = read_csv("datasets/checkin_checkout_history_updated.csv")
plans_data = read_csv("datasets/subscription_plans.csv")

# Clean/tidy 4 datasets
## 1. users_data --> users_tidy:
users_tidy = users_data |>
  mutate(
    name = paste(first_name, last_name),
    membership_days = as.numeric(difftime(as.Date("2024-09-30"), as.Date(sign_up_date, format = "%Y-%m-%d"), units = "days"))
  ) |>
  select(user_id, name, age, gender, membership_days, user_location, subscription_plan)

## 2. gyms_data --> gyms_tidy:
gyms_tidy = gyms_data |>
  mutate(
    climbing_wall = ifelse(grepl("Climbing Wall", facilities), 1, 0),
    swimming_pool = ifelse(grepl("Swimming Pool", facilities), 1, 0),
    basketball_court = ifelse(grepl("Basketball Court", facilities), 1, 0),
    yoga_classes = ifelse(grepl("Yoga Classes", facilities), 1, 0),
    sauna = ifelse(grepl("Sauna", facilities), 1, 0),
    crossfit = ifelse(grepl("CrossFit", facilities), 1, 0)) |>
  select(gym_id, location, gym_type, climbing_wall, swimming_pool, basketball_court, yoga_classes, sauna, crossfit)

## 3. history_data --> history_tidy
history_tidy = history_data |>
  mutate(
    workout_year = year(as.Date(checkin_time, format = "%Y-%m-%d %H:%M:%S")),
    workout_month = month(as.Date(checkin_time, format = "%Y-%m-%d %H:%M:%S")),
    workout_time = format(as.POSIXct(checkin_time, format = "%Y-%m-%d %H:%M:%S"), "%H:%M"),
    workout_duration = as.numeric(difftime(as.POSIXct(checkout_time, format = "%Y-%m-%d %H:%M:%S"),
                                         as.POSIXct(checkin_time, format = "%Y-%m-%d %H:%M:%S"),
                                         units = "mins")),
    workout_timecat = case_when(
      hour(as.POSIXct(checkin_time, format = "%Y-%m-%d %H:%M:%S")) %in% 6:11 ~ "morning",
      hour(as.POSIXct(checkin_time, format = "%Y-%m-%d %H:%M:%S")) %in% 12:17 ~ "afternoon",
      TRUE ~ "evening"
    ),
    calories_per_min = round(ifelse(workout_duration > 0, calories_burned / workout_duration, 0), 2)
  ) |>
  select(user_id, gym_id, workout_year, workout_month, workout_time, workout_timecat, workout_type, workout_duration, calories_burned, calories_per_min)

## 4. plans_data --> plans_tidy
plans_tidy = plans_data |>
  select(subscription_plan, price_per_month)

# Merge 4 datasets to get the final dataset
## 1. users_tidy & history_tidy, by user_id
users_history_tidy = history_tidy |>
  left_join(users_tidy |>
              rename(user_name = name),
            by = "user_id")

## 2. users_history_tidy & gyms_tidy, by gym_id
users_gyms_tidy = users_history_tidy |>
  left_join(gyms_tidy |>
              rename(gym_location = location),
            by = "gym_id")

## 3. users_gyms_tidy & plans_tidy, by subscription_plan
final = users_gyms_tidy |>
  left_join(plans_tidy, by = "subscription_plan")

Importing Raw Datasets

Four raw datasets were imported for analysis:

  • users_data: User demographic and subscription information.

  • gyms_data: Details about gym locations and available facilities.

  • history_data: Workout history, including check-in/out times and calories burned.

  • plans_data: Subscription plans with associated pricing.

Cleaning Steps

  1. users_data:

    • Created name by combining first and last names.

    • Calculated membership_days from the sign-up date to 2024-09-30.

    • Retained essential columns: user_id, name, age, gender, membership_days, user_location, and subscription_plan.

  2. gyms_data:

    • Converted gym facilities into binary indicators (\(1 = present,\ 0 = absent\)) for climbing_wall, swimming_pool, basketball_court, yoga_classes, sauna, and crossfit.

    • Retained relevant columns: gym_id, location, gym_type, and facility indicators.

  3. history_data:

    • Extracted year and month as workout_year and workout_month.

    • Calculated workout_duration (minutes) from check-in and check-out times.

    • Categorized workout times into morning, afternoon, and evening periods (workout_timecat).

    • Calculated calories_per_min as calories burned divided by workout duration.

    • Retained key columns: user_id, gym_id, workout_year, workout_month, workout_time, workout_timecat, workout_type, workout_duration, calories_burned, and calories_per_min.

  4. plans_data:
    • Retained columns for subscription plans and their prices: subscription_plan, price_per_month.

Merging Datasets

  1. Step 1: Merged users_tidy with history_tidy by user_id, renaming name to user_name.

  2. Step 2: Merged users_history_tidy with gyms_tidy by gym_id, renaming location to gym_location.

  3. Step 3: Merged users_gyms_tidy with plans_tidy by subscription_plan.

The final dataset, final, contains 25 variables for analysis and 300000 observations. Below is the codebook for final:

ID Variable Description
1 user_id Unique identifier for each user
2 gym_id ID of the gym where the check-in occurred
3 workout_year Year of check-in time
4 workout_month Month of check-in time
5 workout_time Exact check-in time
6 workout_timecat morning (6:00-11:59); afternoon (12:00-17:59); evening (18:00-24:00)
7 workout_type Type of workout performed during the visit (e.g., Cardio, Weightlifting, Yoga)
8 workout_duration “Check-out time” – “Check-in time” (minute)
9 calories_burned Estimated number of calories burned during the workout
10 calories_per_min Calories burned per minute
11 user_name Users’ full name
12 age Age of the user
13 gender Male; Female; Non-binary
14 membership_days Total days of membership (from signed-up to 2024-9-31–when the datasets were last updated by the author)
15 user_location City where the user lives
16 subscription_plan The user’s gym subscription plan (Basic, Pro, Student)
17 gym_location Location of gym
18 gym_type The type of gym (Premium, Standard, Budget)
19 climbing_wall One of the facilities. 1=yes; 0=no
20 swimming_pool One of the facilities. 1=yes; 0=no
21 basketball_court One of the facilities. 1=yes; 0=no
22 yoga_classes One of the facilities. 1=yes; 0=no
23 sauna One of the facilities. 1=yes; 0=no
24 crossfit One of the facilities. 1=yes; 0=no
25 price_per_month Price per month in Dollar


Exploratory Data Analysis

Calories Burned and Workout Duration

We create a 2D histogram visualizes to represent the relationship between workout duration and calories burned per minute.

plot1 =
  final |>
  ggplot(aes(x = workout_duration, y = calories_per_min)) +
  geom_bin2d() +
  scale_fill_gradientn(colors = c("#FFFFFF", "#A3C9A8", "#F4A261", "#E76F51")) +
  labs(
    title = "Scatter Plot of Workout Duration and Calories Burned", 
    x = "Workout Duration (minutes)", 
    y = "Calories Burned per Minute"
  )
ggplotly(plot1)

It reveals that shorter workout durations are generally associated with higher calorie burn rates, while longer durations show lower burn rates, likely reflecting pacing effects. The highest density of data points appears concentrated at higher durations and moderate calorie burn rates.

Calories Burned and Workout Types

generate_histogram <- function(data, workout_type) {
  plot_ly(
    data = subset(data, workout_type == workout_type),
    x = ~calories_per_min,
    type = "histogram",
    opacity = 0.7,
    name = workout_type
  )
}

workout_type = final |> 
  distinct(workout_type) |> 
  pull(workout_type)

plots = lapply(workout_type, function(wt) generate_histogram(final, wt))

plot2 = subplot(
  plots, 
  nrows = 2,  
  shareX = TRUE, shareY = TRUE  
) %>% layout(
  title = "Calories Burned Per Minute Distribution by Workout Type",
  xaxis = list(title = "Calories Burned Per Minute"),
  yaxis = list(title = "Count")
)

plot2

This histogram shows the distribution of calories burned per minute across different workout types. Most activities cluster around lower calorie burn rates (below 15 calories per minute), with a sharp decline in frequency as the rate increases. The overlap in distributions suggests similarities in calorie burn patterns across workout types, though some types like Yoga and Weightlifting dominate the lower range.

Age and Workout Duration

Age = ggplot(final, aes(x = age)) +
  geom_histogram(
    binwidth = 5
  ) +
  labs(
    title = "Distribution of Age",
    x = "Age",
    y = "Count"
  ) +
  theme_minimal() +
  scale_x_continuous(breaks = seq(20, 70, by = 10)) +
  scale_y_continuous(breaks = seq(0, 35000, by = 10000))+
  theme(plot.title = element_text(hjust = 0.5))
ggplotly (Age)

The age distribution histogram reveals that gym membership is fairly consistent across ages 20-60, with the highest concentration around age 40-45 (~35,000 members). There is a notable decrease in membership after age 60, dropping to about 13,000 members in the 65-70 age group.

Gender and Workout Duration

Gender = ggplot(final, aes(x = workout_duration, fill = gender)) +
  geom_histogram(binwidth = 10) +
  facet_wrap(~gender) +
  labs(title = "Workout Duration by Gender", x = "Workout Duration (minutes)", y = "Count") +
  theme_minimal()+
  theme(plot.title = element_text(hjust = 0.5))
ggplotly(Gender)

Male members show the highest participation (~7,500-8,000 per duration interval), followed by female members (~7,000-7,500), and non-binary members (~2,000). Interestingly, all gender groups display similar workout duration patterns, typically exercising between 30-180 minutes, with consistent numbers in the 60-150 minute range. The similar distribution shapes suggest that workout duration preferences are relatively consistent across genders, with the main difference being in participation numbers rather than duration patterns.

Gym Type and Workout Duration

detailed_stats <- final |> 
  group_by(gym_type) |> 
  summarise(
    mean_duration = mean(workout_duration, na.rm = TRUE),
    median_duration = median(workout_duration, na.rm = TRUE),
    sd_duration = sd(workout_duration, na.rm = TRUE),
    n_users = n()
  )

gym_type <- ggplot(final, aes(x = workout_duration, fill = gym_type)) +
  geom_density(alpha = 0.7) +
  geom_vline(data = detailed_stats, 
             aes(xintercept = mean_duration, color = gym_type),
             linetype = "dashed", size = 1) +
  facet_wrap(~gym_type) +
  scale_fill_manual(values = c("#6C757D",  "#A3C9A8", "#F4A261")) +
  scale_color_manual(values = c("#6C757D",  "#A3C9A8", "#F4A261")) +
  labs(
    title = "Workout Duration Distribution by Gym Type",
    subtitle = "Dashed lines indicate mean duration",
    x = "Workout Duration (minutes)",
    y = "Density"
  )
ggplotly(gym_type)

All three types show similar shaped distributions with no significant skewness, indicating stable exercise habits among all gym members. The only subtle difference is that Premium members tend to maintain slightly longer workout sessions.

Check In Frequency and Workout Duration

user_summary <- final |> 
  group_by(user_id) |> 
  summarize(
    total_check_ins = n(),
    avg_workout_duration = mean(workout_duration, na.rm = TRUE)
  )

Check_in <-ggplot(user_summary, aes(x = total_check_ins, y = avg_workout_duration)) +
  geom_point(alpha = 0.5, color = "#E76F51") +
  geom_smooth(method = "lm", color = "#6C757D") +
  labs(
    title = "Relationship Between Check-In Frequency and Workout Duration",
    x = "Total Check-Ins",
    y = "Average Workout Duration (minutes)"
  ) +
  theme_minimal()+
  theme(plot.title = element_text(hjust = 0.5))

ggplotly(Check_in)

The scatter plot shows no clear relationship between total check-ins and average workout duration, as indicated by the flat trend line. Most users maintain consistent workout durations, clustering around 90–120 minutes, regardless of how often they check in.

Gender with the Highest Workout Duration across Cities

max_duration_by_city <- final |> 
  group_by(gym_location) |>   
  slice_max(order_by = workout_duration, n = 1, with_ties = FALSE) |>   
  select(gym_location, gender)  

max_duration_by_city
## # A tibble: 10 × 2
## # Groups:   gym_location [10]
##    gym_location gender    
##    <chr>        <chr>     
##  1 Chicago      Female    
##  2 Dallas       Male      
##  3 Houston      Female    
##  4 Los Angeles  Male      
##  5 New York     Male      
##  6 Philadelphia Non-binary
##  7 Phoenix      Female    
##  8 San Antonio  Female    
##  9 San Diego    Non-binary
## 10 San Jose     Female
max_duration_by_city |> 
  kable(
    caption = "Gender with the Highest Workout Duration in Each City"
  )|>
    kable_styling(
    full_width = FALSE,
    bootstrap_options = c("striped", "hover", "condensed", "responsive"),
    font_size = 14
  ) |>
  column_spec(1, width = "250px") |> 
  column_spec(2, width = "250px") |> 
  row_spec(0, bold = TRUE, color = "white", background = "#F4A261")
Gender with the Highest Workout Duration in Each City
gym_location gender
Chicago Female
Dallas Male
Houston Female
Los Angeles Male
New York Male
Philadelphia Non-binary
Phoenix Female
San Antonio Female
San Diego Non-binary
San Jose Female

The table highlights the gender associated with the highest workout duration across different cities. Female users dominate, achieving the longest durations in cities like Chicago, Houston, and San Antonio. Male users recorded the highest durations in Dallas, Los Angeles, and New York, while non-binary users had the longest durations in Philadelphia and San Diego.

Subcription Plan by Gender

gender_plan_p =
  users_data |> 
  ggplot(aes(x = subscription_plan, fill = gender)) +
  geom_bar(position = "dodge", width = .6) +
  labs(
    title = "Subscription Plan by Gender", 
    x = "Subscription Plan", 
    y = "Count")+
  theme(plot.title = element_text(hjust = 0.5))

ggplotly(gender_plan_p)

The number of gym-goers choosing the pro subscription plan is similar to the number choosing the basic subscription plan, with slightly more men than women.

Subscription Plan by Cities Where Users Live

city_plan_p =
  users_data |> 
  ggplot(aes(x = user_location, fill = subscription_plan)) +
  geom_bar(position = "dodge", width = .7) +
  labs(
    title = "Subscription Plan by Cities Where Gym-goers Live", 
    x = "City", 
    y = "Count") +
  coord_flip()+
  theme(plot.title = element_text(hjust = 0.5))

ggplotly(city_plan_p)

city_plan =
  users_data |> 
  group_by(user_location, subscription_plan) |> 
  summarize(
    frequency = n()
  ) |> 
  pivot_wider(
    names_from = subscription_plan,
    names_prefix = "plan_",
    values_from = frequency
  )

city_plan |> 
  kable(digits = 3, caption = "Subscription Plan by Cities Where Gym-goers Live")
Subscription Plan by Cities Where Gym-goers Live
user_location plan_Basic plan_Pro plan_Student
Atlanta 156 193 183
Austin 143 177 169
Boston 148 169 142
Denver 169 167 175
Detroit 178 169 168
Las Vegas 181 160 163
Miami 174 162 163
Orlando 161 165 165
San Francisco 166 171 174
Seattle 152 154 183

Atlanta has the most gym-goers choosing pro subscription plan and student plan, and Las Vegas has the most gym-goers choosing basic subscription plan.

Distribution of Age across Subscription Plans

age_plan_p =
  users_data |> 
  ggplot(aes(x = age, fill = subscription_plan)) +
  geom_density(linewidth = .2) + 
  facet_grid(subscription_plan ~ .) +
  labs(
    title = "Distribution of Age across Subscription Plans",
    x = "Age",
    y = "Density"
  )

ggplotly(age_plan_p)

age_plan = 
  users_data |> 
  group_by(subscription_plan) |> 
  summarize(
    age.min = min(age),
    age.max = max(age),
    age.mean = round(mean(age), 0),
  )

The age distribution of gym-goers is relatively even across different subscription plans. The youngest age is 18 years old, the oldest is 64 years old, and the mean age is 41 years old.


Maps

We create an interactive map using the leaflet package to display gym locations and users’ origins along with relevant information in popups.

Gyms’ Locations (in city)

Click on a city to view detailed information about gyms in each location.

city_names = gyms_data |> pull(location)
city_coords = geocode_OSM(city_names)
gyms_data_n = gyms_data

gyms_data_n$lat <- city_coords$lat
gyms_data_n$lon <- city_coords$lon

leaflet(data = gyms_data_n) |> 
  addTiles() |> 
  addMarkers(
    lng = ~lon, lat = ~lat, 
    popup = ~paste(
      "Gym City:", city_names, "<br />",
      "Gym Type:", gym_type, "<br />",
      "Facilities:", facilities)
  )

Gym-goers’ Origins (in city)

Click on a city to view detailed information about the members from each cities.

users_data_n = 
  users_data |> 
  group_by(user_location) |> 
  summarize(
    male.n = sum(gender == "Male"),
    female.n = sum(gender == "Female"),
    age.mean = mean(age),
    basic.n = sum(subscription_plan == "Basic"),
    pro.n = sum(subscription_plan == "Pro"),
    student.n = sum(subscription_plan == "Student")
  )

location_names = users_data_n |> pull(user_location)
location_coords = geocode_OSM(location_names)

users_data_n$lat <- location_coords$lat
users_data_n$lon <- location_coords$lon

leaflet(data = users_data_n) %>%
  addTiles() %>%
  addMarkers(
    lng = ~lon, 
    lat = ~lat, 
    popup = ~paste(
      "User City:", user_location, "<br />", 
      "Average Age:", round(age.mean, 2), "<br />",
      "Male:", male.n, "<br />", 
      "Female:", female.n, "<br />", 
      "Basic Plan Users:", basic.n, "<br />", 
      "Pro Plan Users:", pro.n, "<br />", 
      "Student Plan Users:", student.n
    )
  )


Statistical analysis

Calories Burned Per Minute

We employ descriptive statistical analysis and build a linear regression model to explore the factors influencing calories burned per minute, focusing on workout duration, workout type, demographic, and geographic variables. Key findings highlight the role of workout characteristics, such as shorter durations and specific activities like Swimming, as well as environmental factors like gym facilities and regional differences, in determining calorie-burning efficiency.

Baseline characteristics

final = final |>
  mutate(
    calories_per_min = as.numeric(calories_per_min),
    age = as.numeric(age),
    gender = as.factor(gender),
    workout_month = as.numeric(workout_month),
    workout_timecat = factor(workout_timecat, levels = c("morning", "afternoon", "evening")),
    workout_type = as.factor(workout_type),
    workout_duration = as.numeric(workout_duration),
    membership_days = as.numeric(membership_days),
    user_location = as.factor(user_location),
    subscription_plan = as.factor(subscription_plan),
    gym_type = as.factor(gym_type),
    price_per_month = as.numeric(price_per_month)
  )

mean_calories_per_min = mean(final$calories_per_min, na.rm = TRUE)

final = final |>
  mutate(
    calories_per_min_group = ifelse(calories_per_min > mean_calories_per_min, "High", "Low"),
    calories_per_min_group = as.factor(calories_per_min_group)
  )

vars = c(
  "age", "gender", "workout_month", "workout_timecat", 
  "workout_type", "workout_duration", "membership_days", "user_location", 
  "subscription_plan", "gym_type", "gym_location", "climbing_wall",
  "swimming_pool", "basketball_court", "yoga_classes", "sauna", "crossfit", 
  "price_per_month"
) 

table_one = CreateTableOne(vars = vars, strata = "calories_per_min_group", data = final, test = TRUE)
table_one_df = print(table_one, showAllLevels = TRUE, quote = FALSE, noSpaces = TRUE, printToggle = FALSE)
table_one_df <- table_one_df[, !(colnames(table_one_df) == "test")]

kable(
  table_one_df,
  format = "html",
  caption = "Baseline Characteristics by Calories Burned Group"
) |>
  kable_styling(
    full_width = FALSE,
    bootstrap_options = c("striped", "hover", "condensed", "responsive"),
    font_size = 14
  ) |>
  row_spec(0, bold = TRUE, color = "white", background = "#A3C9A8") |>
  column_spec(2:ncol(table_one_df), width = "3cm") |>
  row_spec(1:nrow(table_one_df), hline_after = TRUE)
Baseline Characteristics by Calories Burned Group
level High Low p
n 107864 192136
age (mean (SD)) 41.23 (13.52) 41.21 (13.55) 0.693
gender (%) Female 46936 (43.5) 83780 (43.6) 0.769
Male 50399 (46.7) 89518 (46.6)
Non-binary 10529 (9.8) 18838 (9.8)
workout_month (mean (SD)) 5.27 (2.74) 5.27 (2.75) 0.889
workout_timecat (%) morning 43192 (40.0) 76698 (39.9) 0.318
afternoon 43096 (40.0) 77288 (40.2)
evening 21576 (20.0) 38150 (19.9)
workout_type (%) Cardio 17831 (16.5) 31798 (16.5) 0.895
CrossFit 17907 (16.6) 32069 (16.7)
Pilates 17984 (16.7) 32051 (16.7)
Swimming 18082 (16.8) 32051 (16.7)
Weightlifting 18175 (16.8) 32131 (16.7)
Yoga 17885 (16.6) 32036 (16.7)
workout_duration (mean (SD)) 71.33 (32.04) 123.15 (37.07) <0.001
membership_days (mean (SD)) 857.96 (295.06) 858.94 (295.43) 0.383
user_location (%) Atlanta 11422 (10.6) 20354 (10.6) 0.840
Austin 10578 (9.8) 18509 (9.6)
Boston 9815 (9.1) 17567 (9.1)
Denver 10992 (10.2) 19658 (10.2)
Detroit 11051 (10.2) 19860 (10.3)
Las Vegas 11001 (10.2) 19352 (10.1)
Miami 10825 (10.0) 19236 (10.0)
Orlando 10534 (9.8) 18967 (9.9)
San Francisco 11052 (10.2) 19729 (10.3)
Seattle 10594 (9.8) 18904 (9.8)
subscription_plan (%) Basic 35210 (32.6) 62523 (32.5) 0.832
Pro 36345 (33.7) 64778 (33.7)
Student 36309 (33.7) 64835 (33.7)
gym_type (%) Budget 32489 (30.1) 57585 (30.0) 0.692
Premium 53812 (49.9) 96075 (50.0)
Standard 21563 (20.0) 38476 (20.0)
gym_location (%) Chicago 11056 (10.2) 19230 (10.0) 0.024
Dallas 10662 (9.9) 19212 (10.0)
Houston 10702 (9.9) 19480 (10.1)
Los Angeles 10625 (9.9) 19270 (10.0)
New York 10853 (10.1) 19099 (9.9)
Philadelphia 10808 (10.0) 19085 (9.9)
Phoenix 10938 (10.1) 19166 (10.0)
San Antonio 10713 (9.9) 19241 (10.0)
San Diego 10625 (9.9) 19310 (10.1)
San Jose 10882 (10.1) 19043 (9.9)
climbing_wall (mean (SD)) 0.50 (0.50) 0.50 (0.50) 0.777
swimming_pool (mean (SD)) 0.70 (0.46) 0.70 (0.46) 0.005
basketball_court (mean (SD)) 0.50 (0.50) 0.50 (0.50) 0.264
yoga_classes (mean (SD)) 0.20 (0.40) 0.20 (0.40) 0.053
sauna (mean (SD)) 0.70 (0.46) 0.70 (0.46) 0.689
crossfit (mean (SD)) 0.40 (0.49) 0.40 (0.49) 0.879
price_per_month (mean (SD)) 26.73 (17.07) 26.73 (17.08) 0.970

The baseline characteristics of the study population, stratified into “High” and “Low” groups based on the mean value of the continuous variable calories burned per minute (calories_per_min), reveal several significant differences. Participants in the low-calorie group exhibited significantly longer workout durations (123.15 ± 37.07 minutes) compared to those in the high-calorie group (71.33 ± 32.04 minutes, p < 0.001). Additionally, gym location showed significant overall variation between the two groups (p = 0.024), suggesting that regional differences may influence workout intensity. Furthermore, the presence of a swimming pool was significantly associated with higher calories burned (p = 0.005), indicating that access to specific facilities may enhance workout effectiveness.

Linear Regression

Build a linear regression model using data-driven model-building process:

\[Y_{calories}=\beta_0 + \beta_1X_{age} + \beta_2X_{gender} + \beta_3X_{user.location} + \beta_4X_{workout.duration} + \beta_5X_{workout.type}+ \beta_6X_{workout.duration}X_{workout.type} + \epsilon\]

model_calories = lm(
  calories_per_min ~ age + gender + user_location + workout_duration + workout_type + workout_duration*workout_type,
  data = final
)
model_calories_summary = tidy(model_calories) |>
  mutate(
    estimate = round(estimate, 5),
    std.error = round(std.error, 5),
    statistic = round(statistic, 2),
    p_value = ifelse(p.value < 0.001, "<0.001", round(p.value, 3))
  ) |>
  select(term, estimate, std.error, statistic, p_value) |>
  rename(
    Variable = term,
    Estimate = estimate,
    `Std` = std.error,
    `T statistic` = statistic,
    `P-value` = p_value
  )

model_calories_summary |>
  kable("html", caption = "Regression Outcome: Calories Burned Per Minute") |>
  kable_styling(
    full_width = FALSE, 
    bootstrap_options = c("striped", "hover", "condensed", "responsive"), 
    font_size = 14,
    position = "center"
  ) |>
  column_spec(1, width = "1cm", extra_css = "text-align: center;") |>
  column_spec(2:5, width = "2.5cm", extra_css = "text-align: center;") |>
  row_spec(0, bold = TRUE, color = "white", background = "#A3C9A8", extra_css = "text-align: center;")
Regression Outcome: Calories Burned Per Minute
Variable Estimate Std T statistic P-value
(Intercept) 23.14789 0.09171 252.41 <0.001
age -0.00047 0.00087 -0.54 0.592
genderMale 0.00839 0.02490 0.34 0.736
genderNon-binary 0.10387 0.04179 2.49 0.013
user_locationAustin 0.09936 0.05249 1.89 0.058
user_locationBoston 0.07072 0.05334 1.33 0.185
user_locationDenver 0.07210 0.05180 1.39 0.164
user_locationDetroit 0.04319 0.05167 0.84 0.403
user_locationLas Vegas 0.06544 0.05190 1.26 0.207
user_locationMiami 0.10592 0.05205 2.03 0.042
user_locationOrlando 0.03679 0.05230 0.70 0.482
user_locationSan Francisco 0.05722 0.05173 1.11 0.269
user_locationSeattle 0.03402 0.05230 0.65 0.515
workout_duration -0.12044 0.00067 -178.79 <0.001
workout_typeCrossFit 0.03685 0.10754 0.34 0.732
workout_typePilates 0.16100 0.10742 1.50 0.134
workout_typeSwimming 0.38898 0.10724 3.63 <0.001
workout_typeWeightlifting 0.29708 0.10708 2.77 0.006
workout_typeYoga 0.23139 0.10734 2.16 0.031
workout_duration:workout_typeCrossFit -0.00024 0.00095 -0.25 0.799
workout_duration:workout_typePilates -0.00128 0.00095 -1.34 0.179
workout_duration:workout_typeSwimming -0.00288 0.00095 -3.04 0.002
workout_duration:workout_typeWeightlifting -0.00217 0.00095 -2.29 0.022
workout_duration:workout_typeYoga -0.00169 0.00095 -1.78 0.075

The linear regression analysis identified several significant predictors of calories burned per minute, revealing the complex interplay of demographic, geographic, and workout-specific factors. Age was not significantly associated with calories burned (p = 0.592), while “Non-binary” gender showed a positive relationship (p = 0.013). Among geographic factors, Miami emerged as a location positively linked to higher calories burned per minute (p = 0.042), whereas other cities, such as Austin, Boston, and Detroit, showed no significant differences. Workout duration exhibited a strong negative association (p < 0.001), indicating that longer workouts tend to result in slightly lower calorie burn rates per minute. Additionally, certain workout types, such as Swimming (p < 0.001), Weightlifting (p = 0.006), and Yoga (p = 0.031), were significantly associated with higher calorie burn rates compared to other activities, highlighting their effectiveness in promoting exercise intensity. However, interaction terms revealed that the positive impact of Swimming (p = 0.002) and Weightlifting (p = 0.022) on calorie burn diminishes slightly as workout duration increases, suggests that calories burn rate of swimming and weightlifting will decrease when workout duration increases (estimates < 0).

Model Validation

We employ cross-validation to compare models with and without interaction terms.

set.seed(123)

calories_df =
  crossv_mc(final, 100) |> 
  mutate(
    train = map(train, as_tibble),
    test = map(test, as_tibble))

calories_df = 
  calories_df |> 
  mutate(
    mod_1  = map(
      train, \(df) lm(calories_per_min ~ age + gender + user_location + workout_duration + workout_type + workout_duration*workout_type, data = df)),
    mod_2  = map(
      train, \(df) lm(calories_per_min ~ age + gender, data = df))
  ) |> 
  mutate(
    rmse_1 = map2_dbl(mod_1, test, \(mod, df) rmse(model = mod, data = df)),
    rmse_2 = map2_dbl(mod_2, test, \(mod, df) rmse(model = mod, data = df))
  )

calories_df |> 
  select(starts_with("rmse")) |> 
  pivot_longer(
    everything(),
    names_to = "model", 
    values_to = "rmse",
    names_prefix = "rmse_") |> 
  mutate(model = fct_inorder(model)) |> 
  ggplot(aes(x = model, y = rmse)) + geom_violin()+
  labs(
    title = "RMSE Distribution Across Models",
    x = "Model",
    y = "Root Mean Squared Error (RMSE)"
  ) +
  theme_minimal()+
  theme(plot.title = element_text(hjust = 0.5) )

The plot suggests that the model with interaction terms has a much lower RMSE, indicating better predictive performance.

Workout Duration

Key Factors Influencing Workout Duration: A Regression Analysis

Build a linear regression model:

\[Y_{workout.duration}=\beta_0 + \beta_1X_{gender} + \beta_2X_{age} + \beta_3X_{user.location} + \beta_4X_{gym.location} + \beta_5X_{subscription.plan}+ \beta_6X_{membership.days} + \beta_7X_{gym.type} + \beta_8X_{gym.facility} + \epsilon\]

final_2 <- final |> 
  mutate(
    gender = as.factor(gender),
    user_location = as.factor(user_location),
    gym_location = as.factor(gym_location),
    subscription_plan = as.factor(subscription_plan),
    gym_type = as.factor(gym_type),
    climbing_wall = as.factor(climbing_wall),
    swimming_pool = as.factor(swimming_pool),
    basketball_court = as.factor(basketball_court),
    yoga_classes = as.factor(yoga_classes),
    sauna = as.factor(sauna),
    crossfit = as.factor(crossfit)
  )

model_duration <- lm(workout_duration ~ 
                  gender + 
                  age + 
                  user_location + 
                  gym_location + 
                  subscription_plan + 
                  membership_days +
                  gym_type + 
                  climbing_wall + 
                  swimming_pool + 
                  basketball_court + 
                  yoga_classes + 
                  sauna + 
                  crossfit,
                  data = final)

model_duration_summary = tidy(model_duration) |>
  mutate(
    estimate = round(estimate, 5),
    std.error = round(std.error, 5),
    statistic = round(statistic, 2),
    p_value = ifelse(p.value < 0.001, "<0.001", round(p.value, 3))
  ) |>
  select(term, estimate, std.error, statistic, p_value) |>
  rename(
    Variable = term,
    Estimate = estimate,
    `Std` = std.error,
    `T statistic` = statistic,
    `P-value` = p_value
  )

model_duration_summary |>
  kable("html", caption = "Regression Outcome: Workout Duration") |>
  kable_styling(
    full_width = FALSE, 
    bootstrap_options = c("striped", "hover", "condensed", "responsive"), 
    font_size = 14,
    position = "center"
  ) |>
  column_spec(1, width = "1cm", extra_css = "text-align: center;") |>
  column_spec(2:5, width = "2.5cm", extra_css = "text-align: center;") |>
  row_spec(0, bold = TRUE, color = "white", background = "#F4A261", extra_css = "text-align: center;")
Regression Outcome: Workout Duration
Variable Estimate Std T statistic P-value
(Intercept) 103.39869 0.49484 208.95 <0.001
genderMale 0.06411 0.16643 0.39 0.7
genderNon-binary 0.62237 0.27933 2.23 0.026
age -0.00340 0.00583 -0.58 0.56
user_locationAustin 0.47724 0.35082 1.36 0.174
user_locationBoston 0.45276 0.35652 1.27 0.204
user_locationDenver 0.77813 0.34627 2.25 0.025
user_locationDetroit 0.42252 0.34546 1.22 0.221
user_locationLas Vegas -0.24860 0.34718 -0.72 0.474
user_locationMiami 0.36621 0.34812 1.05 0.293
user_locationOrlando 0.36606 0.34958 1.05 0.295
user_locationSan Francisco 0.33200 0.34579 0.96 0.337
user_locationSeattle 0.10865 0.34967 0.31 0.756
gym_locationDallas 0.53568 0.35240 1.52 0.128
gym_locationHouston 0.67664 0.35149 1.93 0.054
gym_locationLos Angeles 0.76865 0.35234 2.18 0.029
gym_locationNew York 0.20582 0.35217 0.58 0.559
gym_locationPhiladelphia 0.30881 0.35234 0.88 0.381
gym_locationPhoenix 0.04660 0.35172 0.13 0.895
gym_locationSan Antonio 0.75595 0.35217 2.15 0.032
gym_locationSan Diego 0.59946 0.35223 1.70 0.089
gym_locationSan Jose 0.23524 0.35225 0.67 0.504
subscription_planPro 0.06214 0.19411 0.32 0.749
subscription_planStudent -0.14395 0.19409 -0.74 0.458
membership_days 0.00056 0.00027 2.10 0.036
gym_typePremium NA NA NA NA
gym_typeStandard NA NA NA NA
climbing_wall NA NA NA NA
swimming_pool NA NA NA NA
basketball_court NA NA NA NA
yoga_classes NA NA NA NA
sauna NA NA NA NA
crossfit NA NA NA NA

Age and subscription plan types show no significant impact on workout duration. Non-binary users work out significantly longer than females (estimate = 0.622, p = 0.026), while no significant difference was found between males and females.

Among user locations, Denver residents show significantly longer workout durations (estimate = 0.778, p = 0.025). Gyms in Los Angeles (estimate = 0.769, p = 0.029) and San Antonio (estimate = 0.756, p = 0.032) demonstrate significantly longer workout durations. Membership duration has a small but significant positive effect on workout duration (estimate = 0.001, p = 0.036).

Subscription Plan

In this data analysis, we employ a logistic regression model to assess the relationship between the gym-goers’ subscription plans and the average duration of their workouts, as well as average calories burned and cities where they live.

Subscription Plan and Average duration

factors_plan =
  final |> 
  filter(subscription_plan != "Student") |> 
  mutate(
    plan.pro = ifelse(subscription_plan == "Pro", 1, 0)
  ) |> 
  group_by(user_id, user_location, plan.pro) |> 
  summarize(
    calories.mean = mean(calories_burned),
    duration.mean = mean(workout_duration)
  )

fit_calories = glm(plan.pro ~ calories.mean, 
                   data = factors_plan, family = binomial())

fit_calories |> 
  broom::tidy() |> 
  mutate(OR = exp(estimate)) |>
  select(term, log_OR = estimate, OR, p.value) |> 
  rename(
    Variable = term,
    `P-value` = p.value
  ) |> 
  kable(digits = 3)
Variable log_OR OR P-value
(Intercept) 0.1 1.105 0.855
calories.mean 0.0 1.000 0.906

Subscription Plan and calories burned

fit_duration = glm(plan.pro ~ duration.mean, 
                  data = factors_plan, family = binomial())

fit_duration |> 
  broom::tidy() |> 
  mutate(OR = exp(estimate)) |>
  select(term, log_OR = estimate, OR, p.value) |> 
  rename(
    Variable = term,
    `P-value` = p.value
  ) |> 
  kable(digits = 3)
Variable log_OR OR P-value
(Intercept) -0.095 0.909 0.883
duration.mean 0.001 1.001 0.839

Subscription Plan and cities

fit_cities = glm(plan.pro ~ user_location, 
                    data = factors_plan, family = binomial())

fit_cities |> 
  broom::tidy() |> 
  mutate(
    term = sub("User_location", "", term)) |>
  mutate(OR = exp(estimate)) |>
  select(term, log_OR = estimate, OR, p.value) |> 
  rename(
    Variable = term,
    `P-value` = p.value
  ) |> 
  kable(digits = 3)
Variable log_OR OR P-value
(Intercept) 0.213 1.237 0.048
user_locationAustin 0.000 1.000 0.998
user_locationBoston -0.080 0.923 0.607
user_locationDenver -0.225 0.799 0.143
user_locationDetroit -0.265 0.767 0.082
user_locationLas Vegas -0.336 0.715 0.028
user_locationMiami -0.284 0.753 0.064
user_locationOrlando -0.188 0.828 0.223
user_locationSan Francisco -0.183 0.833 0.232
user_locationSeattle -0.200 0.819 0.203

Overall, Las Vegas is the only city that shows a statistically significant negative impact on subscription plan (p < 0.05), while the effects of other cities are not significant.

Similarly, we are also interested in the relationship between subscription plan and gym types.

Subscription Plan and Gym Types

frequency_plan = 
  final |> 
  group_by(user_id, gym_type, subscription_plan) |> 
  summarize(
    frequency = n()
  )

frequency_plan |> 
  ggplot(aes(x = gym_type, y = frequency, fill = subscription_plan)) +
  geom_boxplot() + 
  labs(
    title = "Frequency of visits by Gym Types",
    x = "Gym Types",
    y = "Frequency"
  )+
  theme(plot.title = element_text(hjust = 0.5) )


frequency_fit = 
  frequency_plan |>
  lm(frequency ~ gym_type, data = _) |> 
  broom::tidy() |> 
  select(term, estimate, std.error, statistic, p.value) |> 
  rename(
    Variable = term,
    Estimate = estimate,
    `Std` = std.error,
    `T statistic` = statistic,
    `P-value` = p.value
  ) 

frequency_fit |> 
  kable(digits = 3)
Variable Estimate Std T statistic P-value
(Intercept) 18.015 0.063 283.934 0
gym_typePremium 11.963 0.090 133.321 0
gym_typeStandard -6.007 0.090 -66.947 0

The results of the linear regression model suggest that gym type significantly influences the frequency of gym visits (p < 0.05).

On average, gym-goers visited the budget gyms 18 times, premium gyms 30 times, and standard gyms 12 times in 2023.


Discussion

Calories Burned Per Minute

The analysis highlights the multifaceted factors influencing calories burned per minute, emphasizing the interplay between workout characteristics, demographics, and environmental factors. Shorter workouts and specific activities, such as Swimming and Weightlifting, are effective in enhancing calorie burn, but their efficiency may decrease as workout duration increases. Geographic and facility-related factors, such as gym location and the presence of swimming pools, further demonstrate the importance of external conditions in shaping workout outcomes.

These findings suggest that tailored exercise plans should consider individual and contextual factors to optimize workout efficiency. By aligning workout duration, activity type, and access to facilities, both personal fitness goals and broader public health outcomes can be better supported. Future research can build on these results to explore strategies that maximize calorie burn and overall exercise effectiveness in diverse populations and environments.

Workout Duration

The age distribution shows consistent gym membership across ages 20-60, with a notable decline after 60. While gender analysis reveals similar workout duration patterns across all groups, non-binary users demonstrate significantly longer workout sessions than other groups. The gym type (Premium/Standard/Budget) shows minimal impact on workout duration, though Premium members tend to maintain slightly longer sessions.

Interestingly, location plays a crucial role, with cities like Denver, Los Angeles, and San Antonio showing significantly longer workout durations. The check-in frequency analysis indicates that members maintain consistent workout durations regardless of visit frequency, suggesting stable exercise habits once established. These findings could be valuable for gym management in optimizing facility operations and developing targeted member engagement strategies.

The data suggests that while demographic factors influence workout patterns, the most significant determinants of workout duration are location-based and membership tenure rather than individual characteristics like age or subscription type. This could inform future decisions about gym locations and membership retention strategies.

Subscription Plan

The results from exploratory data analysis and univariate logistic regression indicate that there is no significant difference in users’ subscription plans by their gender, age, average duration of workouts and average calories burned. It suggests that individual demographic and fitness behavior characteristics may not be the primary drivers of subscription choices. Users’ locations, particularly those in Las Vegas, demonstrate a statistically significant association with subscription plans, highlighting the potential role of geographic and regional factors in influencing membership preferences.

We also found that gym type plays a critical role in determining how frequently users visit the gym. It reflects the impact of facility availability, amenities, or targeted services on user engagement, and helps guide gym management decisions to better meet user needs and preferences.


Conclusion

  • Swimming, weightlifting, and yoga are associated with higher calorie burn efficiency per minute, and shorter, high-intensity workouts are more efficient.

  • Non-binary users and individuals in Miami show higher calorie burn rates, indicating both personal and geographic influences.

  • Non-binary users demonstrate significantly longer workout durations compared to other groups.

  • Location has a strong influence, with cities like Denver, Los Angeles, and San Antonio showing longer workout durations, while gym type has minimal impact.

  • Subscription choices are not significantly influenced by demographics (gender, age) or fitness behaviors (workout duration, calories burned).

  • Gym type plays a crucial role in determining how often users visit the gym.


Shiny App

Calories Burned Calculator is designed to estimate total calories burned during a workout session based on multiple factors such as workout type, duration, gender, age, and location. It incorporates regression models developed by our team, including interaction terms, to provide accurate calorie estimates to help users better understand workout efficiency and trends.

You can find our shiny app here: Calories Burned Calculator

Explore the app to plan your workouts effectively and maximize your calorie burn!